Configure AI Playbooks
Define reusable AI behavior playbooks, manage versions, and control channels and tools from the AI Playbooks page in the Autoch.at app.
What AI Playbooks are
AI Playbooks define structured behavior for your assistants across channels. Each playbook packages a persona, tone, goals, constraints, safety rules, channel settings, and model configuration into a reusable configuration you can apply consistently.
Use playbooks when you need repeatable behavior patterns, for example dedicated flows for lead qualification, customer support, or a general assistant.
AI Playbooks live on an authenticated tenant page. Access depends on a playbooksEnabled feature flag and a canAccessPlaybooks permission check, with admins bypassing the flag. If you do not see AI Playbooks in the sidebar, ask an admin to enable the feature.
URL and navigation
The AI Playbooks page is a dedicated screen under AI configuration.
- Route:
/playbooks - Sidebar: Listed in the main app sidebar under the AI Config section as AI Playbooks.
- Access: Requires you to be signed in to a tenant and to pass the
canAccessPlaybookscheck.
When you open /playbooks, you see the playbooks list on the left and the playbook editor on the right.
Playbooks list and lifecycle
The left-hand panel shows your playbooks in two states: active and archived. Each entry displays:
- Name of the playbook
- Status (for example active vs archived)
- Version label
- Last updated timestamp
From the list, you can:
- Select a playbook to open it in the editor
- Duplicate a playbook to create a starting point based on an existing configuration
- Archive a playbook to remove it from active use without deleting its history
The main editor panel uses a skeleton loader while data is loading, and shows errors and toast messages if operations fail or succeed.
Main workflows
Configure and manage playbooks through a few repeatable workflows.
Create a new playbook
Open the AI Playbooks page at /playbooks and use the New playbook action.
- Enter a clear name for the new playbook so you can recognize it later in the list.
- Choose a template: General Assistant, Lead Qualification, or Customer Support.
The template seeds the Identity, Goals, Tone, and other sections with sensible defaults targeted to that use case. The new playbook appears in the list and opens in the editor as an editable draft.
Edit a playbook and save as draft
With a playbook selected, adjust its behavior in the editor tabs on the right.
- Work through Identity, Tone, Goals, Constraints, Safety, Channels, Tools, Model, and Context as needed.
- Use the Save Draft action to persist your changes without affecting the active behavior of that playbook.
After saving, the editor keeps your changes as the current working draft. Success toasts confirm that the draft saved correctly.
Publish a new version
When your draft is ready for use, publish it as a new version.
- Click Publish New Version.
- Enter a short changelog description so other admins understand what changed in this version.
- Confirm the publish action.
Publishing creates a new version entry for this playbook. The newly published version becomes available in the Versions tab for activation and comparison.
Activate a version
Use the Versions tab in the editor to control which version is currently active.
- In Versions, select a version from the list to view a diff against the currently active version.
- Review changes across the configuration sections in the diff viewer.
- Use Load to bring that version into the main editor if you want to inspect or adjust it further.
- Use Activate to make that version the live configuration used by your assistants.
The active version is clearly labeled so you always know which configuration is in effect.
Archive or duplicate a playbook
From the playbooks list:
- Use Archive to retire a playbook from active use while keeping its history and versions for reference.
- Use Duplicate to copy an existing playbook, including its settings, into a new draft you can adjust independently.
Archiving and duplication both trigger toast notifications with success or error states so you can verify the action completed.
Try changes in the playground
Use the Open Playground action from the editor to test how the current configuration behaves.
- Open Playground while viewing a playbook draft or version.
- Interact with the assistant using the configured Identity, Tone, Goals, Constraints, and other settings.
- Adjust the configuration and reopen Playground as needed until responses meet your expectations.
Use the playground before publishing a new version to reduce surprises once the configuration goes live.
Editor sections
The right-hand editor is organized into tabs so you can reason about different parts of behavior separately while still managing a single playbook.
Use Identity to define the base persona and prompt that shape how the assistant sees itself.
- Describe who the assistant is, who it helps, and how it should introduce itself.
- Capture any persistent role instructions, such as acting as a sales rep, support agent, or onboarding guide.
Identity is the foundation for downstream behavior; adjust this first when you build a new playbook.
The Tone tab controls how responses sound.
- Choose overall tone settings, such as formal vs conversational.
- Add custom tone rules, like avoiding jargon or using short paragraphs.
- Use tone overrides for specific types of content where needed.
Tone works alongside Identity to make responses feel consistent with your brand.
In Goals, define what success looks like for this playbook.
- Set a primary goal, for example qualifying a lead, resolving a support ticket, or booking a demo.
- Add secondary goals or success criteria, such as capturing contact details or confirming next steps.
Clear goals help the model prioritize actions and structure conversations toward outcomes rather than just answering questions.
Use Constraints to limit behavior and output.
- Add never rules that specify actions or content the assistant must avoid.
- Configure maximum response length to keep messages concise, especially on constrained channels like SMS.
Constraints are useful for regulatory requirements, brand safety, and keeping responses on-task.
The Safety tab defines how the assistant escalates and handles risky situations.
- Configure escalation triggers, such as mentions of self-harm, harassment, or sensitive topics.
- Define handling policies, describing when to involve a human, what to say, and what information to collect.
Safety policies keep responses aligned with your risk tolerance and compliance needs.
The Channels tab lets you adapt behavior per communication channel, such as email, SMS, chat, and voice.
- Toggle each channel on or off for this playbook.
- Add channel-specific prompt additions or instructions.
- Override tone per channel if, for example, email should be more formal than chat.
- Set maximum length per channel so SMS stays within character limits while email responses can be more detailed.
Channel settings help you reuse the same core playbook while still respecting medium-specific constraints.
In Tools, control which tools the assistant can use.
- Maintain lists of allowed tools for this playbook.
- Mark tools as denied to block use in certain playbooks.
- Require approval for specific tools, so high-impact actions do not occur without explicit confirmation.
Tool controls let you align capabilities with each playbook’s purpose and risk profile.
The Model tab configures the underlying LLM behavior.
- View the provider configured for the tenant; this appears disabled in the UI for this page.
- Select the model to use if multiple options are available for your provider.
- Adjust temperature to control creativity vs determinism.
- Set maximum tokens to cap the size of responses and manage costs.
Model settings fine-tune how strictly the assistant follows instructions and how verbose it is.
Use Context to connect the playbook to tenant data and reference material.
- Toggle inclusion of products and website content.
- Attach specific document IDs so the assistant can reference targeted documents.
- Add custom instructions that provide extra background or constraints tied to your content.
Context makes the playbook grounded in your real data instead of only generic knowledge.
The Versions tab manages historical and active versions of the playbook.
- Browse a version list, each with metadata such as label or timestamp.
- Use the diff viewer to compare any version with the active version across Identity, Tone, Goals, and other sections.
- Load a version into the editor to inspect or tweak it.
- Activate a version to make it the current live configuration for this playbook.
Version control lets you iterate safely, roll back quickly, and keep a clear audit trail of changes.
Best practices for admins and advanced users
When you configure AI Playbooks for a tenant, start with templates to move quickly, then refine.
- Map each playbook to a clear use case, such as a specific team or funnel stage.
- Keep Identity and Goals tightly aligned; adjust Constraints and Safety before rolling out widely.
- Use the Playground to validate behavior after any meaningful change, then publish a new version with a descriptive changelog.
- Prefer duplicating an existing playbook over editing a heavily used one if you are experimenting with major changes.
Related pages
Use these pages together with AI Playbooks to control end-to-end AI behavior.
Last updated 2 weeks ago
Built with Documentation.AI